Having dealt with similar problems on the implementation side it's incredible how in this day and age it's still so difficult to pass state through sign-in or sign-up.
Half of the reason appears to be how the Oauth 2.0 standard is written - it purposefully forbids browser-specific items like the hash fragment in the URL. For passing state you have the "state" field but, despite the name, its original purpose was to prevent XSRF attacks, not store data.
Vendor libraries don't go to huge lengths to help here, as they settle on complying with the standard.
True, but it's extremely effective at getting registrations, almost to the point that you're dooming yourself without it. Anyone and everyone uses it, big and small. Google can afford any random user to run a few queries before dangling the hook, but it's all the same.
Ux on mobile is very bad. First you lure me with a prompt, but when I type it it's lost and I'm redirected to login page. Ok, I signed up, logged in. On mobile the progress panel is 1cm wide, I didn't notice and I set the same prompt now I have 2 generating and no way to cancel. You (correctly) started title of this post with the word "generate" that is your core feature, everything else I see on mobile is unrelated bullshit, including the carousel, including the "choose your role" dialog. On mobile there should be prompt (50% of screen real estate), a generate button, output/progress (50% of screen real estate) and hamburger button that hides everything else.
The generation haven't finished yet, I will report when it finishes to see if core feature is even usable.
Mobile is tricky, since such design software are very difficult to use on a small mobile screen. We are looking into it though. How was the generation?
Update: on second try the generation succeeded, the output was hallucination, completely unrelated to the prompt. Useless. Image comparing what I wanted and what was generated: https://files.catbox.moe/kplzpk.png
Thank you for the feedback! we will consider them both for the website and the product. We have several examples in the product, we will push them to be present on the webpage
I've been doing a lot of 3D design in Codex GPT 5.5 (I found Opus 4.7 wasn't as good - haven't experimented much with 4.8 or Fable).
OpenSCAD is a parametric CAD programming language, and the models know it well.
The biggest challenge is communicating words like "inside" and "above" to the model - inevitably it's idea of which direction is which is often different.
I can't say I've done anything very hard, but for things like ESP32 cases, or parametric rod connectors it is great.
You can do things like "add snap connectors" and it'll do a great job.
Tried two very open ended prompts: "Home robot" and "Jet engine". First one returned only two plates connected with 4 rods, second model did not produce any 3d model at all.
Curious whether some hierarchical modeling could be applied here.
Examples guys. You are telling me a lot and not showing me. Get rid of the home page animation that does nothing and replace it with examples that show real world use cases, step by step.
Would be curious to hear from folks who do modelling:
My first reaction from seeing the examples on the page, is that they are somewhat simple, almost standardized, or come from templates. They also don't look very time-consuming to model yourself. The product advertises itself on the parameterization aspect, but my hunch is that practitioners have their personal library of models which can be rescaled without much work. Is that somewhat accurate?
36 comments
[ 2.4 ms ] story [ 35.5 ms ] threadMy few hundred word prompt wasn't saved/persisted. :'(
Half of the reason appears to be how the Oauth 2.0 standard is written - it purposefully forbids browser-specific items like the hash fragment in the URL. For passing state you have the "state" field but, despite the name, its original purpose was to prevent XSRF attacks, not store data.
Vendor libraries don't go to huge lengths to help here, as they settle on complying with the standard.
This bad implementation of it makes it even worse.
I hate it with a passion and I will boycott services that do those inconsiderate things to me.
I’ve tried Gemini (paid) in the past with openscad and then designed the same object myself and I was 10 times faster
The generation haven't finished yet, I will report when it finishes to see if core feature is even usable.
Still generating, 53 minutes.
Why would you use 50% of the screen space when you can have 100% and flip between prompt and render?
And I'm not going to sign up just to test it.
I agree with the other ones, have 20 cached examples or at least some videos that show the process from start to finish.
Show us the results. Don't talk about the future. Everyone can yield empty promises "in the future". That means nothing.
OpenSCAD is a parametric CAD programming language, and the models know it well.
The biggest challenge is communicating words like "inside" and "above" to the model - inevitably it's idea of which direction is which is often different.
I can't say I've done anything very hard, but for things like ESP32 cases, or parametric rod connectors it is great.
You can do things like "add snap connectors" and it'll do a great job.
Curious whether some hierarchical modeling could be applied here.
No thanks.
My first reaction from seeing the examples on the page, is that they are somewhat simple, almost standardized, or come from templates. They also don't look very time-consuming to model yourself. The product advertises itself on the parameterization aspect, but my hunch is that practitioners have their personal library of models which can be rescaled without much work. Is that somewhat accurate?
A library/toolkit for the Swift programming language to built parametric 3d objects.
Give Claude access to the docs and examples and you can do amazing things.